In [3]:
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
from sklearn.metrics import mean_squared_error, mean_squared_log_error
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import GRU, Dense
# Generate a simple dataset
np.random.seed(42)
time = np.arange(0, 100, 0.1)
sin_wave = np.sin(time) + 0.5 * np.random.randn(len(time))
# Create a DataFrame
data = pd.DataFrame({'timestamp': time, 'sin_wave': sin_wave})
# Split data into training and testing sets
train_size = int(0.8 * len(data))
train_data = data[:train_size]
test_data = data[train_size:]
# List of different window sizes to test
window_sizes = [5, 10, 15, 50, 100]
# Function to create dataset for GRU
def create_dataset(dataset, time_steps=1):
dataX, dataY = [], []
for i in range(len(dataset) - time_steps):
a = dataset['sin_wave'].values[i:(i + time_steps)]
dataX.append(a)
dataY.append(dataset['sin_wave'].values[i + time_steps])
return np.array(dataX), np.array(dataY)
# Initialize arrays to store results
all_mse = []
all_rmse = []
# Build a common GRU model
def build_model(input_shape):
model = Sequential()
model.add(GRU(units=10, return_sequences=False, input_shape=input_shape))
model.add(Dense(units=1))
model.compile(optimizer='adam', loss='mean_squared_error')
return model
# Create a subplot for plotting
fig, ax = plt.subplots(len(window_sizes), 1, figsize=(12, 6 * len(window_sizes)))
# Train and evaluate GRU model for different window sizes
for idx, time_steps in enumerate(window_sizes):
X_train, y_train = create_dataset(train_data, time_steps)
X_test, y_test = create_dataset(test_data, time_steps)
# Reshape input to [samples, time_steps, features]
input_shape = (X_train.shape[1], 1)
X_train = np.reshape(X_train, (X_train.shape[0],) + input_shape)
X_test = np.reshape(X_test, (X_test.shape[0],) + input_shape)
# Build and train the model
model = build_model(input_shape)
model.fit(X_train, y_train, epochs=50, batch_size=64, verbose=0)
# Predict using the trained model
predictions = model.predict(X_test)
# Calculate evaluation metrics
mse = mean_squared_error(y_test, predictions)
rmse = np.sqrt(mse)
# Store the calculated MSE and RMSE
all_mse.append(mse)
all_rmse.append(rmse)
# Plot the forecast
ax[idx].plot(test_data['timestamp'].values[time_steps:], y_test, label='Actual')
ax[idx].plot(test_data['timestamp'].values[time_steps:], predictions, label='Forecasted', linestyle='dashed')
ax[idx].set_xlabel('Timestamp')
ax[idx].set_ylabel('Gas Concentration')
ax[idx].set_title(f'Window Size: {time_steps}')
ax[idx].legend()
# Print evaluation metrics for each window size
for idx, time_steps in enumerate(window_sizes):
print(f"Window Size: {time_steps}")
print("Mean Squared Error:", all_mse[idx])
print("Root Mean Squared Error:", all_rmse[idx])
# Adjust layout and show plots
plt.tight_layout()
plt.show()
print("layer = 2 , units = 50\n\n")
# Initialize arrays to store results
all_mse = []
all_rmse = []
# Build a common GRU model
def build_model(input_shape):
model = Sequential()
model.add(GRU(units=50, return_sequences=True, input_shape=input_shape))
model.add(GRU(units=50, return_sequences=False))
model.add(Dense(units=1)) # 1 output unit
model.compile(optimizer='adam', loss='mean_squared_error')
return model
# Create a subplot for plotting
fig, ax = plt.subplots(len(window_sizes), 1, figsize=(12, 6 * len(window_sizes)))
# Train and evaluate GRU model for different window sizes
for idx, time_steps in enumerate(window_sizes):
X_train, y_train = create_dataset(train_data, time_steps)
X_test, y_test = create_dataset(test_data, time_steps)
# Reshape input to [samples, time_steps, features]
input_shape = (X_train.shape[1], 1)
X_train = np.reshape(X_train, (X_train.shape[0],) + input_shape)
X_test = np.reshape(X_test, (X_test.shape[0],) + input_shape)
# Build and train the model
model = build_model(input_shape)
model.fit(X_train, y_train, epochs=50, batch_size=64, verbose=0)
# Predict using the trained model
predictions = model.predict(X_test)
# Calculate evaluation metrics
mse = mean_squared_error(y_test, predictions)
rmse = np.sqrt(mse)
# Store the calculated MSE and RMSE
all_mse.append(mse)
all_rmse.append(rmse)
# Plot the forecast
ax[idx].plot(test_data['timestamp'].values[time_steps:], y_test, label='Actual')
ax[idx].plot(test_data['timestamp'].values[time_steps:], predictions, label='Forecasted', linestyle='dashed')
ax[idx].set_xlabel('Timestamp')
ax[idx].set_ylabel('Gas Concentration')
ax[idx].set_title(f'Window Size: {time_steps}')
ax[idx].legend()
# Print evaluation metrics for each window size
for idx, time_steps in enumerate(window_sizes):
print(f"Window Size: {time_steps}")
print("Mean Squared Error:", all_mse[idx])
print("Root Mean Squared Error:", all_rmse[idx])
# Adjust layout and show plots
plt.tight_layout()
plt.show()
# Define the naive model function
def naive_model(test_data):
y_test_naive = np.roll(test_data, shift=1)
y_test_naive[0] = test_data[0] # Replace the first value with the original value
return y_test_naive
# Apply the naive model to test data
y_test_naive = naive_model(test_data['sin_wave'].values)
# Plot the results for both sensors
plt.figure(figsize=(12, 6))
plt.plot(test_data['timestamp'].values[1:], y_test_naive[1:], label='Predicted - Naive Model', linestyle='dashed')
plt.plot(test_data['timestamp'].values[1:], test_data['sin_wave'].values[1:], label='Actual')
plt.xlabel('Timestamp')
plt.ylabel('Gas Concentration')
plt.title('Actual and Predicted Gas Concentration (Naive Model with Operational Shift One)')
plt.legend()
plt.grid(True)
plt.show()
# Calculate MSE for the Naive Model predictions
mse_naive = mean_squared_error(test_data['sin_wave'].values[1:], y_test_naive[1:])
rmse_naive = np.sqrt(mse_naive)
print(f"MSE value for Naive Model: {mse_naive:.4f}")
print(f"RMSE value for Naive Model: {rmse_naive:.4f}")
7/7 [==============================] - 0s 1ms/step 6/6 [==============================] - 0s 2ms/step 6/6 [==============================] - 0s 2ms/step 5/5 [==============================] - 0s 3ms/step 4/4 [==============================] - 0s 5ms/step Window Size: 5 Mean Squared Error: 0.30140706494034974 Root Mean Squared Error: 0.5490055235973038 Window Size: 10 Mean Squared Error: 0.3012241272667031 Root Mean Squared Error: 0.5488388900822382 Window Size: 15 Mean Squared Error: 0.290877647984537 Root Mean Squared Error: 0.539330740811737 Window Size: 50 Mean Squared Error: 0.3008202319895981 Root Mean Squared Error: 0.5484708123406369 Window Size: 100 Mean Squared Error: 0.3001316976371685 Root Mean Squared Error: 0.5478427672582422
layer = 2 , units = 50 7/7 [==============================] - 1s 2ms/step 6/6 [==============================] - 1s 3ms/step 6/6 [==============================] - 1s 4ms/step 5/5 [==============================] - 1s 8ms/step 4/4 [==============================] - 1s 15ms/step Window Size: 5 Mean Squared Error: 0.30017545439102317 Root Mean Squared Error: 0.5478827013066055 Window Size: 10 Mean Squared Error: 0.29812383423934236 Root Mean Squared Error: 0.5460071741647196 Window Size: 15 Mean Squared Error: 0.28177928268045593 Root Mean Squared Error: 0.5308288638350932 Window Size: 50 Mean Squared Error: 0.2548915166072231 Root Mean Squared Error: 0.5048678209266492 Window Size: 100 Mean Squared Error: 0.2571802286835185 Root Mean Squared Error: 0.5071294003343905
MSE value for Naive Model: 0.4458 RMSE value for Naive Model: 0.6677
